Financial contagion during crises: A spatial approach for CDS spreads

Financial contagion during crises: A spatial

approach for CDS spreads

by

Gerjan Halmingh 1927833

A thesis submitted to University of Groningen in conformity with the requirements for the degree of Master of Science.

May, 2016

MSc Economics MSc Finance

2

Financial contagion during crises: A spatial

approach for CDS spreads

Gerjan Halmingh

May 2016

Abstract

This paper analyses financial contagion of banks in the Eurozone during crisis periods from 2007 till 2012. As a proxy for the default risk of banks, CDS spreads are used. We modify Merton's structural model of default by replacing the risk free rate with sovereign CDS spreads and add a spatial interaction effect of the CDS spreads of banks. By adding a spatial interaction term, the idiosyncratic risk and systemic risk components are separately analysed. First, we find that all variables of the idiosyncratic risk component are in line with previous research. For the systemic risk component, a spatial weight matrix is constructed on the basis of international trade as a proxy for interbank linkages. We find significant empirical

evidence in favour of increasing spatial interaction effects during times of crisis. These excess spatial interaction effects determine that there is contagion in bank CDS spreads and provides evidence of a more macroprudential regulatory approach. The results are robust with respect differences in the spatial weight matrix.

JEL Classification Numbers: G01, G12, G15,G21, C31

3

1. Introduction.

In August 2007 the French bank BNP Paribas publicly announced that they could not "fairly value" the collateralized debt obligations (CDO’s) they had on their balance sheet, which can be viewed as a basket of credit default swaps (Longstaff and Rajan, 2008). A credit default swap (CDS) is a derivative based on credit risk transfers which can be regarded as an insurance protecting credit claims. The insurance premium that the protection buyer pays is referred to as the CDS spread, which are often times measured as basis points. Though BNP Paribas is but mere one example, it reflects the increasing uncertainty in financial markets. Due to interbank linkages, the difficulties faced by BNP Paribas, and many other financial institutions for that matter, spreaded worldwide and what followed were multiple bank bailouts and the worst global financial crisis in decades. Partly because of moral hazard, not every bank that ran into financial trouble was bailed out and on September 15, 2008, Lehman Brothers collapsed. With Lehman holding more than $691 billion in assets, this became the largest bankruptcy in history and triggered a major shock for the global economy and created a domino effect of default for multiple banks. The banks that were bailed out had vastly increased the burdens on national debt (IMF,2009) which eventually started the European sovereign debt crisis. On October 19, 2009, the Greek Prime Minister George Papandreou, publicly revealed that Greece encountered severe fiscal difficulties. In the coming years, it became clear that multiple other countries in the Eurozone, most notably Ireland, Italy, Portugal and Spain (together the so-called GIIPS countries), had accumulated unstable debt levels as well. In 2009, Greece had a deficit of 12,7% which the Stability and Growth

4 the outstanding debt. The ECB further decreased overall default risk on September 6, 2012 when they announced a precautionary bailout programme for all Eurozone countries (ECB, 2012). Though there are no clear cut-off points as to when the financial crises started or when it ended, there is a general consensus that it started somewhere in the summer of 2007 and ended late 2008 (IMF, 2013). The paper uses the public announcement of the French bank BNP Paribas as the start of the crisis and the end of 2008 as its occlusion. By that time, markets slowly stabilised again until minister George Papandreou revealed the earnestness of Greek’s fiscal difficulties and heralded the European sovereign debt crisis which lasted till the bailout programme of September 2012. The European sovereign debt crisis starts and ends with the announcements of George Papandreou and the ECB, respectively (Arellano et al., 2012).

1.1. Goal of the paper

Though there has been much microprudential regulation and supervision, most notably by setting the minimum capital requirements as mentioned in the Basel agreements, the global financial crisis has shown us that these should be enhanced by a macro perspective in order to reduce systemic risk, which refers to situations where multiple financial institutions fail as a result of a common shock (Allen et al., 2010). This papers aims to find results that will aid macroprudential regulation and supervision in order to lessen future financial crises. The focus of this paper lies on the contagion of the global financial crisis and the European sovereign debt crisis. Contagion is defined by Forbes and Rigobon (2002) as a significant increase in cross-market linkages after a shock to one or multiple countries. This paper follows the definition of Bekaert et al. (2005) who elaborates on Forbes and Rigobon and identifies financial contagion as excess cross-country correlation of asset prices during crisis periods. In order to measure the financial contagion during the crises, spatial spillovers of the increasing overall default risk are used. Spatial spillovers are used for two reasons. First of all, they represent the pure negative externalities of financial contagion (Capello, 2009).

Secondly, due to the cross-sectional dependence of the data, standard statistical estimators are biased and inconsistent while spatial econometric techniques are robust. Altogether, the main question this paper addresses, is in what way there exists financial contagion between the default risk of banks in the Eurozone during the financial- and sovereign debt crises of August 2007 until August 2012. To measure default, CDS spreads are used, which following

5 spatial spillovers measure the indirect effects of an increase in the CDS spread of one

financial institution on another financial institution. This effect is expected to be positive during crises, that is, that there exist externalities in the form of increased systemic risk. Although this has been researched more often, the main difference lies in the methodology used. A dynamic spatial spillover model is used in order to distinguish between the systemic and idiosyncratic risk components of financial institutions over time. It estimates CDS spreads of financial institutions based on their idiosyncratic risk determinants and provides estimates for the spatial spillovers of the systemic risk component in order to analyse the financial contagion. Knowing this can improve regulation and supervision in a more macroprudential way.

The remaining sections of the paper are structured as follows: Section 2 provides a literature review, Section 3 introduces the methodology and Section 4 discusses the data. Section 5 show the results which are further examined in the robustness test of Section 6. Section 7 concludes.

2. Literature review

There has already been a tremendous amount of research on financial crises due to their impact. As stated by Allen and Gale (2000, p.2) "financial crises are important because they raise the costs of intermediation and restrict credit, which in turn restrain the level of activity in the real sector and ultimately can lead to periods of low growth and recession". Crises can spread over from one area to the other due to systemic risk which is what happened in the United States: what started in a small part in the United States went global due to high systemic risk (Eichengreen et all., 2012) and the worldwide contagion was a fact. Rigobon (2002) states that since the Asian crisis of 1997, contagion is one of the most debated topics in international finance. Evidently, financial contagion has been estimated by various

6 This paper aims to shed light upon the dispersion of default risk in times of crisis.

Understanding credit risk of the banking sector is of utmost importance, for bankruptcies of financial institutions impose severe externalities on the real economy (Acharya et all., 2010). There are several indicators of bank and sovereign credit risk for which the most

straightforward are CDS spreads and bond yields for banks and their sovereignty. Aizenman et al. (2013) state three main advantages in favour of CDS spreads. First of all, CDS spreads indicate a more timely market-based pricing of risk. Furthermore, by using bond yields, a problem arises concerning the time to maturity, which CDS spreads overcome. Last but not least, bond yields pricing include multiple dimensions, that is, it incorporates inflation expectations, a lending based supply and demand state as well as default risk. Since the main focus of this paper concerns default risk, CDS spreads are used. The main hypothesis of the paper is that spatial spillovers for CDS spreads of banks increase during times of crisis. While the main variable of interest is the CDS spreads of banks, it is impossible to separate this from the CDS spreads of sovereignties. Due to regulatory requirements, banks have a relatively large exposure to sovereign debt (Popov and van Horen, 2014), implying that the stability of banks and the sovereignty are positively correlated. This dependency of stability works the other way around as well, countries can decrease the risk of banks whenever they absorb (part of) the debt. That is, the reduction in the financial market risk is traded for an increase in sovereign risk (Acharya, Drechsler and Schnabl, 2014). This is shown empirically by several authors. Eising and Lemke (2011) who show that bailouts decrease the CDS spreads of banks while simultaneously increases sovereign CDS spreads. Alter and Schuler (2012) find similar results but further state that after the bailout, increases in sovereign default risk increases CDS spreads of banks as well. Avino and Cotter (2014, p.84) state that, "especially during crisis periods, sovereign CDS spreads may incorporate more timely information on the default probability of European banks than their corresponding bank CDS spreads." De Bruyckere et al. (2013) show that contagion occurred between bank and sovereign CDS spreads in Europe during the 2007-2012 period. There is in general a positive relation between CDS spreads of banks and their sovereignty. However, when the government absorb (part of) the debt of distressed banks, this relationship becomes negative. As for the sovereign default risk, Gerlach et al. (2010) indicate that the size of the banking sector is of importance as well.

2.1 Idiosyncratic risk component

7 by many. Merton's (1974) structural model of default provides us with a theoretical

framework that encompasses leverage, volatility and the risk-free rate. Though already more than 40 years have passed since its publication, Merton’s path breaking paper is still the workhorse for gaining insights about capital structure (Sundaresan, 2013). As shown by Benkert (2004), this theory applies to CDS pricing and Ericsson et al. (2009) also find that leverage, volatility and the risk-free rate are important determinants of CDS spreads. As a proxy for the risk free rate, government bonds are generally used. Kagraoka and Moussa (2014) have shown that there is strong correlation between government bonds and sovereign CDS spreads. This paper builds upon the insights of Merton’s structural model of default and its application towards CDS pricing by two means. First, due to strong correlation of

government bonds and sovereign CDS spreads, and the timely information that sovereign CDS spreads incorporate on the default probability of European banks, we re-estimate Merton’s model while replacing the risk free rate with sovereign CDS spreads. Secondly, we test for spatial interaction effects between the CDS spreads of banks in order to test for financial contagion. Following Merton, we expect that the idiosyncratic part of CDS spreads, that is, leverage and volatility, have a positive effect on CDS spreads. While sovereign CDS spreads have a mixed effect on bank CDS spreads, we follow Popov and van Horen (2014) in that stability of banks and their sovereignty are positively correlated. This implies that we overall expect a positive impact of sovereign CDS spreads on bank CDS spreads.

3. Methodology

Since the aim of the paper is the financial contagion of the European sovereign debt crisis, data of 33 banks located in the Eurozone are used. We limit the geographical dispersion of the data used to only the Eurozone for two reasons. First, in order to accurate estimate the model, we do not want interference of currency exchange rates or debt devaluation and therefore stick to the European monetary union. Secondly, the ECB established a precautionary bailout programme for all Eurozone countries, therefore creating a discrepancy between the Eurozone and other regions. As for time boundaries, we collect data from August 2007, the

8 Germany have 4 and Italy and Spain have 5. For the remaining two countries, Luxembourg and Slovenia, no data was available of banks for the full time period.

To measure the risk of banks, the 5 year CDS spreads as a mid-price in basis points of the contract’s notional value are used as the endogenous variable. 5 year CDS spreads are taken for they are the most liquid and other CDS spreads may have pricing errors (Kagraoka and Moussa, 2014). Daily CDS spreads of all 33 banks in the Eurozone are used for the period of August 2007 until September 2012. Since CDS spreads are market driven, and markets are closed during the weekends, the total time periods used for the 61 months from August 2007 till September 2012 is 1328 days. This paper uses panel data with N = 33 and T = 1328. Due to the relation between bank CDS spread and its sovereignties' CDS spread, the daily 5 year CDS spread as a mid-price in basis points of each of the 13 countries in the Eurozone is also used. As for the determinants of the idiosyncratic risk component of the CDS spreads, the paper partly follows Merton's structural model, implying that leverage and volatility are used as the exogenous variables. For volatility, daily backward looking historical data are used which measures the volatility in the stock prices of banks over the last month and leverage is measured as the debt/equity ratio as a percentage on a daily basis. Since the focus of this paper lies on financial contagion during crises, a dummy variable is added that equals 1 for a crisis, and 0 otherwise, where the periods of August 2007 till December 2009 and from 19 October onwards are defined as crises.

3.1 Ordinary Least Squares

In order to test whether or not Merton's structural model with the modification of replacing the risk free rate with sovereign CDS spreads still holds, we estimate a fixed effects ordinary least squares (OLS) model as in equation 1.

with (1)

9 volatility and sovereign CDS. Since we cannot assume that the explanatory variable are fully independent of the unit specific error terms, a fixed effects model is used rather than a random effects model. Following the empirical work of (among others) Merton, we expect that the coefficients of the idiosyncratic risk component of CDS spreads, that is β1 and β2, are positive.

Likewise, following Popov and van Horen (2014) we expect that the coefficient β3 which

measures the effect of sovereign CDS spreads on bank CDS spreads is positive as well. Formally stated:

HA: β1 > 0, HA: β2 > 0, HA: β3 > 0

H0: β1 = 0, H0: β2 = 0, H0: β3 = 0

3.2 Spatial autoregressive model

The analysis incorporates an endogenous interaction effect which measures the interaction between CDS spreads of a bank a with the CDS spreads of a bank b and vice versa. The line of thought is that, after controlling for idiosyncratic risk and government default risk, the relation between the CDS spreads of banks a and b measure financial linkages of default in the multinational model. The model is estimated by the following spatial autoregressive model (SAR):

(2)

where N = 33 and T = 1328. Yt is a column vector of 33 x 1 representing CDS spreads of the

33 banks at time t, Xt is a 33 x 3 matrix which include the idiosyncratic risk component of

banks as well as the sovereign default risk, where each element xi,k corresponds to the kth

10 CDS spreads of banks are market priced and thus exhibit behavioral financial imperfections such that market prices do not fully reflect the true fundamental prices. Prior to the crisis periods, investors are overly optimistic, thereby driving asset prices of default below the true fundamental price. As soon as a crises unfolds, investors’ expectations are adjusted such that it reflects less optimism, driving the asset prices of default, that is, the CDS spreads of banks, upwards (Emine, 2009). We expect to be zero for non-crisis periods and non-zero for crisis periods. Formally:

H0: = 0,

HA: > 0,

where indicates the estimated rho’s during normal (non-crisis) times and indicates the estimated rho’s during crisis times. Since the main focus of the paper lies on financial

contagion, which Forbes and Rigobon identified as excess correlation in times of crisis, merely estimating does not suffice. The estimates of only identifies spatial correlation and interbank linkages, not financial contagion. In order to test for financial contagion, must be higher than it would normally be during non-crisis days. This is tested by equation 3.

+ (3)

where is a 1328 x 1 vector with the estimated rho’s from equation 2, one estimate rho for each day. D is a 1328 x 1 vector which contains a dummy variable which equals 1 for a crisis and 0 otherwise. Since the main hypothesis states that spatial interaction in times of crisis increase, that is, that there exists financial contagion, we expect that delta is positive:

HA:

H0:

11 which we expect, just as in the case of the fixed effects OLS model, to be positive for

leverage, volatility and the sovereign CDS spread. Formally stated:

HA: > 0, HA: > 0, HA: > 0

H0: = 0, H0: = 0, H0: = 0

where measure the effect of leverage, volatility and sovereign CDS spread on the CDS spread of banks, respectively.

3.4. Direct and indirect spatial spillovers

Though equation 2 provides useful information as to whether or not there exist spatial

spillovers, it does not comprise the picture as a whole and may lead to erroneous conclusions (LeSage and Pace, 2009). By analysing the direct and indirect spillovers of the explanatory variables we gain more sophisticated knowledge about overall spatial spillovers. While the dynamic spatial autoregressive model as described by equation 2 provides unbiased and consistent statistical estimators, its major drawback lies in these direct and indirect spatial spillovers. The direct and indirect spillovers are calculated by equation 4.

(4)

where I is the identity matrix and corresponds the betas estimated by equation 2 where k = 1, 2, 3 corresponds to leverage, volatility and sovereign CDS spreads, respectively. The direct effect of the explanatory variables are calculated as the mean of the diagonal elements

whereas the indirect effect of the explanatory variables are calculated as the mean row sum of the off-diagonal elements. The drawback of the SAR used in equation 2 is that due to the appearance of in both the numerator and the denominator, the ratio between the indirect and the direct effect of the explanatory variables is independent of which implies that the magnitude only depends on the SAR parameter and the spatial weight matrix W (Elhorst, 2014). Though this implication will most likely not hold empirically many times we

12

4. Data

4.1. The spatial weight matrix

Since we are dealing with spatial dependence, the main question involves how the spatial weight matrix must be constructed. It goes without saying that the interbank linkages in the Eurozone go far beyond than the immediate neighbours of countries and that banking market in the Eurozone is reasonably integrated. However, Kleimeier et al. (2013) argue that, despite banking market integration, geographical dispersion is still the key determinant of cross-border lending. Though this favours a geographical spatial weight matrix, many argue

differently. Claessens et al. (2010), Blanchard et al (2010) and Cetorelli and Goldberg (2011) all state that integration with global financial markets plays a major role. Since export and import of countries increases with their trade integration while taking into account the geographical distance as important barrier, the spatial weight matrix is, as a proxy for

interbank linkages, constructed on the basis of exports between countries and their GDP. All values of exports and GDP will be taken before the crisis so that the spatial weight matrix does not become endogenous.

The spatial weight matrix W is an N×N matrix which describes the spatial composition of the banks. N equals the number of banks so that W is a 33×33 matrix. The spatial composition is measured by exports between countries and their GDP. The relatedness of banks in the spatial weight matrix, before row normalisation, is given by equation 5.

(5)

where Wi,x,j,y represents the relatedness between bank i in country x and bank j in country y

(for all i ≠ j) and are the elements of matrix W, Ex,y is the export from country x to country y

and Ny is the number of banks in country y. GDPx represents GDP in country x, E is total

exports from country x and Nx is the number of banks in country x. The spatial compositions

are divided by Nx or Ny due to otherwise over representativeness of dependence, that is, if

equation 5 did not divide by Nx or Ny, it implies that if there are more banks in county y, total

13 interbank linkages would be huge with respect to foreign banks if the number of banks

increases in the domestic country while not correcting for this. After the spatial weight matrix is created, row normalisation is used as described by equation 6 where wij is the weight

associated with each element in matrix W. It follows by definition that for i = j, wij = 0.

(6)

Though there is a lot of trade between the 13 countries, trade theory tells that trade diminishes with geographical distance. Though we only incorporate data from 13 Eurozone countries, increasing the number of countries with countries with higher geographical distance has a diminishing marginal effect on the total trade making the elements of the matrix grow less fast. Therefore, row and column sums of the spatial weight matrix (before row-normalisation) does not diverge to infinity at a rate equal to or faster than the number of banks, which is a sufficient condition for the spatial weight matrix (Lee, 2004). This ensures that correlation between spatial units converge to zero as the distance goes to infinity.

4.2 Descriptive statistics

14

Figure 1: CDS spreads of banks in GIIPS countries vs. those in non-GIIPS countries

15

Figure 2: CDS spreads of GIIPS Countries vs non-GIIPS countries.

16

Figure 3: Average volatility of banks in GIIPS countries vs non-GIIPS

Table 1 provides the correlation between variables. Though in general most are not

particularly high, the correlation between bank CDS and sovereign CDS is over 50%. This derives from the theory of Popov and van Horen (2014) who state that due to regulatory requirements, banks have a large exposure to sovereign debt.

Bank CDS Leverage Volatility Sovereign CDS

Bank CDS 1.000 0.083 0.116 0.511

Leverage 0.083 1.000 0.014 0.070

Volatility 0.116 0.014 1.000 0.023

Sovereign CDS 0.511 0.070 0.023 1.000

Table 1: Correlation among variables

17 high for our time sample, by the end of 2007 financial institutions had tripled their leverage when compared to the late 1990s (IMF, 2009).

Bank CDS Leverage Volatility Sovereign CDS

Mean 279.45 27.65 19.10 121.71

Std. Error 396.35 161.49 100.12 180.35

Coefficient of variation 0.71 0.17 0.19 0.67

Table 2: Descriptive analysis of variables

5. Results

5.1 Ordinary least squares

Though the main focus of the paper lies on spatial interaction effects, the model presented in equation 2 would not be very useful if the non-spatial component is not explained by

leverage, volatility and the sovereign CDS spreads. Using fixed effects ordinary least squares we find that CDS spreads of banks are indeed explained by the non-spatial variables. As can be seen from table 3, all coefficients are significant with p-values of 0. The variables describe bank CDS spreads quite well with an adjusted of more than 36%. The results support the theory, all coefficients are positive and significant.

Coefficient Std. Error t-Statistic P-value

Leverage 0.152 0.009 16.11 0.00 Volatility 0.571 0.015 37.56 0.00 Sovereign CDS spread 0.742 0.010 77.36 0.00 Constant 173.960 1.920 90.63 0.00 0.388 Adjusted 0.369 Prob F-statistic 0.00

18 5.2 Spatial Spillovers

Since Merton's structural model (with risk free rate replaced by sovereign CDS spreads) works, the spatial autoregressive model is tested. We estimated equation 2 for all 1328 time periods. Table 4 reports the average statistics of the 1328 regressions. It shows that the main variable of interest rho, which measures spatial interaction, is positive but not significant. This non-significance follows from behavioral financial theory of investors’ optimism during times of non-crisis and the fact that CDS spreads are market driven. Figure 4 plots the estimated rho's over time in which we can see that rho during crisis periods differs from that of non-crisis periods. Furthermore, we see that rho is positive during the European sovereign debt crisis while it is not clearly so during the financial crisis of 2007-2008. Figure 4 clearly shows that when George Papandreou on October 2009 announced Greece's unstable debt levels, spatial interaction of bank CDS spreads started to increase rapidly. There was a small decrease in spatial interaction when the Stability and Growth Programme was declared in January 2010. At January 2012, when Standard & Poor downgraded many sovereignties, rho started to increase rapidly again until the announcement of the ECB on March 2012 that it activates the buy back scheme put a stop to this. After the announcement, spatial interaction between bank CDS spreads started to decrease towards zero.

Coefficient Std. Error T-statistic P-value

Rho 0.083 0.058 1.43 0.15 Leverage 1.353 0.916 1.48 0.14 Volatility 0.362 0.134 2.70 0.01 Sovereign CDS 267.680 98.687 2.71 0.01 Constant -391.737 -154.263 2.54 0.01 0.566 Adjusted 0.521

Table 4: Spatial output from equation 2

19 anymore. The adjusted increased with respect to the OLS model in equation 1 with more than 15%, implying that, though not statistically significant, spatial spillovers might indeed enhance the model.

Figure 4: Estimated rho's over time.

Since the focus of the paper lies on financial contagion, that is, excess correlation of asset prices during crisis periods, mere positive estimators for rho do not explain financial

contagion. These spillovers should be higher during crisis periods. To test whether increases in rho coincides with crisis periods as figure 4 suggest, equation 3 is performed using standard OLS. Table 5 reports the outcome of the regression. As it shows, the coefficient for the

dummy variable is positive and significant with a p-value of 0. The positive coefficient shows that during crises, spatial interaction between CDS spreads of banks increases, that is, there is financial contagion.

Coefficient Std. Error t-Statistic P-value

Dummy 0.122 0.007 17.61 0.00 Constant -0.014 0.006 -2.19 0.03 0.180 Adjusted 0.179 Prob F-statistic 0.00

20 5.4 Direct effects of exogenous variables

The positive delta as estimated by equation 3 states that there is indeed financial contagion during crisis periods. However the estimated rho's over time as plotted in figure 4 suggest that the significance of delta is mainly attributable to the European sovereign debt crisis rather than both crises. By analyzing the direct and indirect effects we can clearly find whether there exist contagion in the financial crisis of 2007-2008 as well. Table 6 reports the average

statistics estimated by equation 4. By looking at the direct effects we find that all coefficients are still positive such as what was expected. However, leverage and volatilities are no longer statistically significant on average for the whole time sample.

Coefficient Std. Error T-statistic p-value

Direct Leverage 1.402 1.045 1.34 0.18

Direct Volatility 0.400 0.310 1.29 0.20

Direct Sovereign CDS spread 2.793 0.033 84.61 0.00

Indirect Leverage 0.381 0.265 1.44 0.15

Indirect Volatility 0.342 0.159 2.15 0.03

Indirect Sovereign CDS spread 1.067 0.008 138.24 0.00

Table 6: output for direct and indirect effects

21 volatility on bank CDS spreads is positive but not significant as shown in table 6. Figure 7 shows the direct effect of sovereign CDS spreads on bank CDS spreads over time. During the early stages of the financial crisis of 2007-2008, its direct effect remained stable which lasted until the fall of the Lehman Brothers on September 15 2008. By that time, the direct effect of sovereign CDS spreads temporarily spiked but stabilized again at the end of 2008 which is used as the occlusion of the financial crisis. During the start of the European sovereign debt crisis, the direct effect of sovereign CDS spreads slowly started to rise, shortly stabilized after the announcement of the Stability and Growth Programme and then continued to grow until the announcement of Greece's central bank that the government cash deficit was reduced by more than 40%. After that, the direct effect of sovereign CDS spreads temporarily declined and then spiked right before the first announcement of the ECB. After the second

announcement of the ECB, the direct effect of sovereign CDS spreads decreased again. Over the entire time sample, the direct effect of sovereign CDS spreads is significant with a p-value of 0.

22

Figure 6: Direct effect of volatility

Figure 7:Direct effect of sovereign CDS spreads

5.5 Indirect effects of exogenous variables

23 the indirect effects of leverage are clearly positive. We see that the indirect effect of leverage stabilizes again after the announcement of the ECB on March 2012. This discrepancy of indirect effects during both crises holds for volatility as well. Figure 9 clearly shows no significant indirect impact of volatility during the financial crisis of 2007-2008 while showing clear deviations from 0 during the European sovereign debt crisis. It shows that these

deviations started on October 19 2009 where George Papandreou revealed the earnest of Greece's fiscal difficulties. Just as in figure 6, there exists a large drop of the effect of volatility during the downgrading of Standard & Poor. Furthermore, the indirect effect of volatility stabilizes, just like leverage, after the ECB announcement. Overall, the indirect spillovers from the volatility of banks are positive for the sample as a whole at the p = 0,05 level. Not surprisingly, the indirect effect of sovereign CDS spreads rose tremendously during the European sovereign debt crisis as shown in figure 10. While stable during the financial crisis of 2007-2008, the indirect spillovers started to increase rapidly after October 19 2009 and decreased and stabilized after the announcement of the ECB on March 2012. Overall, even by including the financial crisis of 2007-2008, the indirect spillover effect of sovereign CDS spreads remains highly significant with a p-value of 0.

24

Figure 9: Indirect effect of volatility

Figure 10: Indirect effect of sovereign CDS spreads

6. Robustness Test

Since we are dealing with a spatial autoregressive model, the chosen specification of W is of vital importance (Leenders, 2002). In order to test whether the results are robust with respect to the specification of the spatial weight matrix, we re-estimate the model by using a

25 matrix are calculated by equation 7. Since geographical data is used and not all countries are neighbors of each other, the row and column sums of matrix W (before row-normalization) are bounded in absolute value, which is a sufficient condition for the spatial weight matrix (Kelejian and Prucha, 1999) and ensures that the correlation between spatial units converge to zero as the distance goes to infinity. After the construction of matrix W, row normalization is used such as describes by equation 2 where again wij = 0 for all i = j.

(7)

26 Coefficient Std. Error T-statistic P-value

Rho 0.032 0.030 1.09 0.28 Leverage 1.475 0.990 1.49 0.14 Volatility 0.513 0.195 2.63 0.01 Sovereign CDS 315.092 88.884 3.54 0.00 Constant -461.631 -146.375 3.15 0.00 0.545 Adjusted 0.498

Table 7: Spatial output from equation 2 using the geographical matrix

Figure 11: Difference in Rho’s

27 Coefficient Std. Error T-Statistic P-Value

Dummy 0.206 0.012 17.54 0.00

Constant -0.011 0.010 -1.10 0.27

0.188

Adjusted 0.188

Prob F-statistic 0.000

Table 8: output from equation 3 using the geographical spatial weight matrix.

7. Conclusion

28 in time can clearly be seen. These results were robust for different forms of spatial weight matrices.

The paper gives strong evidence that solely microprudential regulation such as the Basel agreements is not sufficiently enough to obviate following financial crises. Financial

contagion occurs through spatial spillovers, which this paper has shown increases more than normally during financial crisis. In order to prevent or lessen future financial crises, policies should be more macroprudential-focussed, limiting the excess spillover effects during crisis and ensure no domino effect happens. As markets become more integrated, the issue of interbank linkages and their potentially large problems are at the heart of policymakers who should be aware that mere microprudential regulation does not suffice anymore.

7.1 Limitations and further research

Though the results of the paper are robust for differences in the spatial weight matrix, a main issue remains the limited set of banks that were used. Whereas the large time span of the paper provides data for 1328 days, the number of banks is limited to 33. The main regression analysis of the paper provides estimates of a spatial autoregressive model that merely contains 33 units. Though all estimates support the hypothesis, the results can be more robust if N grows. As for overall robustness, the model stands or falls with the specification of the spatial weight matrix. While both matrices W and W2 _{based on export between countries and}

29

8. List of references

Acharya, V., Drechsler, I., Schnabl, P. 2014, "A pyrrhic victory? Bank bailouts and sovereign credit risk." The Journal of Finance, 69(6), 2689-2739.

Acharya, V., Pedersen, L., Philippon, T., Richardson, M. 2010, “Measuring Systemic Risk.” Working Paper, New York University

Aizenman, J., Hutchison, M., Jinjarak, Y. 2013, "What is the risk of European sovereign debt defaults? Fiscal space, CDS spreads and market pricing of risk." Journal of International Money and Finance 34 (1), 37–59.

Allen, F., Gale, D. 2000, "Financial Contagion." Journal of Political Economy 108(1), 1-33.

Allen, F., Carletti, E. 2010, "An Overview of the Crisis: Causes, Consequences, and Solutions." International Review of Finance. Vol. 10, Issue 1, p1-26. 26p

Alter, A., Schuler, Y. 2012, "Credit spread interdependencies of European states and banks during the financial crisis." Journal of Banking and Finance 36 (12), 3444–3468.

Andrikopoulos, A., Samitasa, A., Kougepsakisb, K. 2014, " Volatility transmission across currencies and stock markets: GIIPS in crisis." Applied Financial Economics, Vol. 24, Issue 19, 1261–1283

Arellano, C., Conesa, C., Kehoe, T. 2012, " Chronic Sovereign Debt Crises in the Eurozone, 2010-2012." p4-11. 8p.

Avino, D. Cotter, J. 2014, "Sovereign and Bank CDS Spreads: Two Sides of the Same Coin?" Journal of International Financial Markets, Institutions and Money, Vol. 32, pp. 72-85

30 Bekaert, G,, Campbell R. Harvey, and Angela Ng, “Market Integration and Contagion,” Journal of Business 78 (2005):39–69.

Benkert, C. 2004, "Explaining Credit Default Swap Premia." Journal of Futures Markets, Vol. 24, Issue 1, pp. 71-92

Blanchard, O., Das, M., Faruqee, H. 2010, "The Initial Impact of the Crisis on Emerging Market Countries." Brookings Papers on Economic Activity, Issue 1, p263-323. 61p

De Bruyckere, V., Gerhardt, M., Schepens, G., Vander Vennet, R. 2013, "Bank/sovereign risk spillovers in the European debt crisis." Journal of Banking & Finance

Capello, R. 2009, “Spatial spillovers and regional growth: a cognitive approach.” European planning studies , Vol. 17, Issue 5, p639-658. 20p

Cetorelli, N. & Goldberg, L. 2011, "Global Banks and International Shock Transmission: Evidence from the Crisis." IMF Economic Review, Vol. 59, Issue 1, pp. 41-76

Claessens, S., Dell’Ariccia, G., Igan, D., Laeven, L. 2010, "Cross-country experiences and policy implications from the global financial crisis." Economic Policy, Vol, 25 Issue 62

ECB 2012, "Technical features of Outright Monetary Transactions", ECB Press Release

Eder, A., Keiler, S. 2015, "CDS spreads and contagion amongst systemically important financial institutions - a spatial economic approach." International Journal of Finance and Economics 20: 291–309

Edwards, S., Susmel, R. 2001, "Volatility dependence and contagion in emerging equity markets." Journal of Development Economics 66(2): 505–532.

31 Ejsing, J., Lemke, W. 2011, "The Janus-headed salvation: sovereign and bank credit risk premia during 2008-09." Economic Letters 110 (1), 28–31.

Elhorst, P. 2014, “Spatial Econometrics: From Cross-Sectional Data to Spatial Panels.” Springer Briefs in Regional Science. New York and Heidelberg: Springer

Emine, B. 2009, “Can Miracles lead to Crises? The role of Optimism in Emerging Markets Crises.” Journal of Money, Credit & Banking, Vol 41 Issue 6, p1189-1215. 27p.

Ericsson, J., Jacobs, K., Oviedo, R. 2009, "The determinants of credit default swap premia." Journal of financial and quantitative analysis." Vol. 44, Issue 1, pp, 109-132

Forbes, K., Rigobon, R. 2002, "No contagion, only interdependence: Measuring stock market comovements." Journal of Finance, 5

Gerlach, S., Schulz, A., Wolff, B. 2010, " Banking and sovereign risk in the euro area" Deutsche Bundesbank, Research Centre, Discussion Paper Series 1: Economic Studies: 09

IMF 2009, "Fiscal implications of the global economic and financial crisis." IMF Staff Position Note.

IMF 2013, "Financial Crises: Explanations, Types, and implications." IMF Staff Position Note.

Kagraoka, Y., Moussa, Z. 2014, "Estimation of the Term Structure of CDS-Adjusted Risk-Free Interest Rates." Journal of Fixed Income, Vol. 24 Issue 2, p29-44. 16p

Kelejian, H., Prucha, I. 1999, " A generalized moments estimator for the autoregressive parameter in a spatial model." International Economic Review, Vol. 40 Issue 2, p509. 24p

32 Lee, L. 2004, "Asymptotic distributions of quasi-maximum likelihood estimators for spatial autoregressive models." Econometrica, Vol. 72, Issue 6, pp. 1899-1925

Leenders, R., 2002, "Modeling social influence through network autocorrelation: constructing the weight matrix." Social Networks, Vol 24, Issue 1, PII S0378-8733(01)00049-1, pp. 21-47.

LeSage, J., Pace, R. 2009, “Introduction to spatial econometrics.” CRC Press, Taylor & Francis Group, Boca Raton

Longstaff, F., Mithal, S., Neis, E. 2005, "Corporate Yield Spreads: Default Risk or Liquidity? New Evidence from the Credit Default Swap Market." Journal of Finance, Vol. 60 Issue 5, p2213-2253. 41p

Longstaff, F., Rajan, A. 2008, "An Empirical Analysis of the Pricing of Collateralized Debt Obligations." Journal of Finance, Vol. 63 Issue 2, p529-563. 35p

Merton, R. C. 1974, "On the Pricing of Corporate Debt: The Risk Structure of Interest Rates." Journal of Finance, 29 , 449^70.

Popov, A., Van Horen, N. 2014, "Exporting sovereign stress: Evidence from syndicated bank lending during the euro area sovereign debt crisis." Review of Finance forthcoming.